Regional predictions of sub-daily rainfall extremes through data-driven blends of morphoclimatic descriptors

Due to the limited length of locally available sequences of precipitation extremes, estimates of design rainstorms at a given location (i.e. point rainfall depth associated with given durations and non-exceedance probabilities) are traditionally obtained from regional frequency analysis. Several statistical regionalization methods proposed in the literature enable one to exploit sequences of precipitation extremes observed at a number of sites that supposedly share the same frequency regime of rainfall extremes with the site of interest (herein also referred to as a homogeneous pooling group of sites). Homogeneous pooling groups of sites can be identified by looking at specific climatic descriptors; for instance, some reliable authors successfully utilize Mean Annual Precipitation (MAP) as the sole proxy for locally characterizing the frequency regime of sub-daily rainfall extremes and for grouping sequences of rainfall extremes records. We aim at advancing this traditional approach (1) by relaxing the hypothesis of the existence of a homogeneous pooling group of sites characterized by a unique regional parent distribution and (2) by incorporating additional morphological and climatic information in the regional model. Our research focuses on a large study area in Northern Italy, counting more than 2350 Annual Maximum Series of rainfall depth for different time-aggregation intervals between 1 and 24 hours, that have been collected between 1928 and 2011 in the Italian Rainfall Extreme Dataset (I2-RED).  We refer to local MAP value as well as to several other morphologic descriptors (e.g. minimum distance to the coast, elevation of orographic barriers, aspect, terrain slope, etc.) for characterizing the frequency regime of sub-daily rainfall extremes. We train a probabilistic neural network that uses the descriptors cited above as input layers for modeling the local frequency regime of observed rainfall annual maxima. We resort to a Generalized Extreme Value (GEV) distribution whose parameters are data-driven functions of the local morphoclimatic descriptors as well as the time-aggregation interval. We then perform a series of cross-validation experiments targeted at assessing the accuracy of the developed data-driven regional frequency model relative to a simpler regional model in which GEV parameters are functions of MAP and time aggregation intervals.

Our results address the following research problems: (a) identification of the most descriptive morphological proxies for representing the frequency regime of sub-daily rainfall extremes, (b) assessment of potential, limitations, and robustness of data-driven multivariate regional frequency models of sub-daily rainfall extremes relative to simpler and more traditional regionalization schemes.